on the volatility and comovement of u.s. financial markets...
TRANSCRIPT
On the Volatility and Comovement of U.S. Financial
Markets Around Macroeconomic News Announcements
Menachem Brenner, Paolo Pasquariello, and Marti Subrahmanyam1
January 30, 2008
1Stern School of Business, New York University, Ross School of Business, University of Michigan, and
Stern School of Business, New York University, respectively. Please address comments to the authors at
[email protected], [email protected], and [email protected]. A previous version
circulated with the title “Financial Markets and the Macro Economy.” We benefited from the comments
of an anonymous referee, Robert Engle, Amit Goyal, Clara Vega, Guojun Wu, and participants in
seminars at University of New South Wales, University of Melbourne, the 2007 AFA meetings, Monash
University, and Bank of Korea. Any remaining errors are our own.
Abstract
The objective of this paper is to provide a deeper insight into the links between financial markets
and the real economy. To that end, we study the short-term anticipation and response of U.S.
stock, Treasury, and corporate bond markets to the first release of surprise U.S. macroeconomic
information. Specifically, we focus on the impact of these announcements not only on the level,
but also on the volatility and comovement of those assets’ returns. We do so by estimating several
extensions of the parsimonious GARCH-DCC model of Engle (2002) for the excess holding-
period returns on seven portfolios of these asset classes. We find that both the process of
price formation in each of those financial markets and their interaction appear to be driven by
fundamentals. Yet, our analysis reveals a statistically and economically significant dichotomy
between the reaction of the stock and bond markets to the arrival of unexpected fundamental
information. We also show that the conditional mean, volatility, and comovement among stock,
Treasury, and corporate bond returns react asymmetrically to the information content of these
surprise announcements. Overall, the above results shed new light on the mechanisms by which
new information is incorporated into prices within and across U.S. financial markets.
JEL classification: C32; G14
Keywords: Volatility; Comovement; Information; Macroeconomic News
Abstract
The objective of this paper is to provide a deeper insight into the links between financial markets
and the real economy. To that end, we study the short-term anticipation and response of U.S.
stock, Treasury, and corporate bond markets to the first release of surprise U.S. macroeconomic
information. Specifically, we focus on the impact of these announcements not only on the level,
but also on the volatility and comovement of those assets’ returns. We do so by estimating several
extensions of the parsimonious GARCH-DCC model of Engle (2002) for the excess holding-
period returns on seven portfolios of these asset classes. We find that both the process of
price formation in each of those financial markets and their interaction appear to be driven by
fundamentals. Yet, our analysis reveals a statistically and economically significant dichotomy
between the reaction of the stock and bond markets to the arrival of unexpected fundamental
information. We also show that the conditional mean, volatility, and comovement among stock,
Treasury, and corporate bond returns react asymmetrically to the information content of these
surprise announcements. Overall, the above results shed new light on the mechanisms by which
new information is incorporated into prices within and across U.S. financial markets.
JEL classification: C32; G14
Keywords: Volatility; Comovement; Information; Macroeconomic News
1 Introduction
There seems to be little doubt among academics and practitioners that the release of macroe-
conomic news has a significant impact on prices of securities within as diverse asset classes as
stocks, Treasury bonds, and corporate bonds. For instance, most asset pricing models provide
a snapshot of the cross-sectional relationship between asset returns (or prices) and risk factors
at a given point in time. A change in one or more of these factors should, therefore, affect
asset returns, with the dynamic nature of these changes being dictated by the dynamics of new
information arriving in the market.
The dominant paradigm regarding the response of asset prices to new information (first artic-
ulated by Fama, 1971) is that, since markets are efficient, asset prices should react immediately
and in an unbiased manner to new information. Accordingly, in this paper we explore the func-
tioning of the process of price formation in all three of the main U.S. financial markets – stocks,
government bonds, and corporate bonds – around the release dates of the unexpected portion
of four important macroeconomic news: the Consumer Price Index (CPI), the Unemployment
Rate, the Nonfarm Payroll Employment, and the Target Fed Funds Rate. Specifically, we intend
to address four basic sets of questions.
First, what is the impact of these macroeconomic announcements on asset returns and asset
return volatility in the proximity of their first release? Are the markets where these assets are
traded more volatile before the news event and less volatile afterward? This would be the case,
for example, if the arrival of information leads to resolution of uncertainty and/or disagreement
among market participants (e.g., Pasquariello, 2007). However, the analysis of Ross (1989) may
lead to the opposite inference. Ross argues that, in an arbitrage-free economy, the volatility of
prices should be related to the arrival of information in an efficient market. Accordingly, Foster
and Viswanathan (1993) and Pasquariello and Vega (2007) show that, ceteris paribus, both the
availability and the realizations of a public signal increase price volatility.
Second, does the release of macroeconomic news surprises affect the markets for different
asset classes in different ways? The rationale for heterogeneity in the fluctuations of stock and
(government and corporate) bond returns in response to macroeconomic news is intuitive. For
instance, higher inflation may increase only the volatility of the bond market, since stocks (and,
to a lesser extent, corporate bonds) offer a natural hedge against inflation.
Third, do macroeconomic announcements affect the existing degree of covariance between
different asset classes? Covariance shifts may stem from information spillovers across markets
(e.g., Shiller,1989; King and Wadhwani,1990; Pindyck and Rotemberg, 1990, 1993; Karolyi and
Stulz, 1996; Connolly and Wang, 2000), from more intense portfolio rebalancing activity across
1
stocks and bonds (e.g., Fleming et al., 1998; Kodres and Pritsker, 2002), from financial con-
straints (e.g., Kyle and Xiong, 2001), or from greater dispersion of beliefs among speculators
(e.g., Kallberg and Pasquariello, 2007; Pasquariello, 2007) before and after their arrivals.
Fourth, is the impact of unexpected macroeconomic news releases instantaneous or protracted
over time? Are the observed shocks to volatility and comovement of a transitory or persistent
nature? For instance, Jones et al. (1998) observe that clustering of information arrivals, market
sentiment, or gradual learning may justify why an increase in return volatility would persist
over time. Therefore, lack of persistence in announcement shocks would suggest that additional
information gathering or the trading process do not intrinsically increase return volatility.1
To tackle these questions, we combine a database of daily excess holding-period returns on
seven asset portfolios – stocks traded on the NYSE-AMEX and NASDAQ, five, ten, and thirty-
year Treasury bonds, and Aaa and Baa-rated long-term corporate bonds – with data on those
four macroeconomic announcements and their consensus expectations between 1986 and 2002.
The choice of daily spot data is not casual. First, higher-frequency data are unavailable for two
of the asset classes under consideration – corporate and government bonds – over our entire
sample period. Since one important aspect of our study is to compare the reaction of all U.S.
financial markets to the release of U.S. macroeconomic news, full sample coverage for all asset
classes represents a necessary prerequisite. Second, using higher-frequency data over shorter
sample periods, recent studies find a significant intraday reaction of some of these markets to
macroeconomic news, often within minutes of their release (e.g., Balduzzi et al., 2001; Andersen
et al. 2003, 2005; Green, 2004; Fleming and Piazzesi, 2005). Thus, the daily frequency of our
data is most likely to bias our estimates of that reaction downward. Yet, as we soon discuss, our
evidence is instead strong. Lastly, higher-frequency data are afflicted by microstructure frictions
(e.g., bid-ask bounce, quote clustering, price staleness, inventory effects) that may bias inferences
drawn upon them (e.g., Bai et al., 2000; Andersen, Bollerslev, and Meddahi, 2005 and references
therein). These frictions generally become immaterial over longer horizons.2
The expectation data – from the International Money Markets Services survey (for macroe-
conomic variables) and from futures prices (for Fed Funds rate decisions) – are customarily
assumed to represent unbiased estimates of the anticipated portion of these announcements.3
Hence, they allow us to identify the unexpected component in those information arrivals when
released to the public. We then estimate the impact of these events on conditional returns, return
1Jones et al. (1998) provide some supporting evidence for this argument in the U.S. Treasury market using
actual (rather than unexpected) employment and producer price announcements.2E.g., see the discussion in Hasbrouck (2006).3The main properties of these datasets and references to their use in the literature are given in Section 2.
2
volatility, and return covariance using several extensions of the GARCH(1,1)-DCC IMA model
of Engle (2002). This specification, albeit more flexible and parsimonious than most available
multivariate models, has been shown to perform equally well in a variety of situations.
The objective of our study is to contribute to the empirical literature relating macroeconomic
fundamentals to asset pricing dynamics. It is argued there that the impact of news arrivals
on asset prices occurs through any of three channels: expected cash flows from the assets, the
appropriate discount rates, and their risk premiums (e.g., see Andersen et al., 2005 for a review).
In particular, the development of asset pricing models including macroeconomic risk factors dates
back to the Intertemporal Capital Asset Pricing Model (ICAPM) of Merton (1973), but it was
the empirical analysis of Chen et al. (1986) that inspired further investigations of the effect
of macroeconomic variables on asset returns. Chen et al. (1986) argue that surprise shocks to
macroeconomic factors (such as inflation and industrial production) should affect the discount
rate and the expected cash flows of individual firms and, in turn, their stock returns. Yet, their
econometric approach yields only mixed results. This is not surprising since these factors may
affect each of those three primitive factors (e.g., see Boyd et al., 2005), leaving the net effect
ambiguous. Consequently, many of the ensuing studies concentrate explicitly on the impact of
various macroeconomic news announcements on asset prices.4
These studies differ on multiple grounds: their choice of news, their choice of market (bonds,
stocks, or currencies), the moments of the return distribution they examine, and the statistical
methodology they employ.5 Only a few of these studies focus on the simultaneous impact of
macroeconomic news releases on both stocks and government bonds or their futures and none
on the corporate bond market. Some of these studies use the actual news releases, i.e., do not
separate the expected component of the released information from the unexpected one (e.g., Jones
et al., 1998; Fleming and Remolona, 1999). Most of these studies concentrate on the first and
second moments of asset returns. Fleming et al. (1998) consider the volatility linkages of stocks,
bonds, and money market instruments, while Connolly et al. (2005, 2007) investigate the impact
of stock uncertainty (measured by the implied volatility from equity index options) on the time-
varying comovement between government bond and stock returns within and across countries.
Yet, neither of these studies specifically relates the estimated spillovers to macroeconomic news.
4Some recent studies also analyze the impact of macroeconomic news releases on stock and bond market
liquidity (e.g., Brandt and Kavajecz, 2004; Green, 2004; Vega, 2006; Pasquariello and Vega, 2007).5An incomplete list of the most recent research includes Jones et al. (1998), Li and Engle (1998), Fleming
and Remolona (1999), Bomfim and Reinhart (2000), Connolly and Wang (2000), Balduzzi et al. (2001), Kuttner
(2001), Flannery and Protopapadakis (2002), Andersen et al. (2003, 2005), Bomfim (2003), Bernanke and Kuttner
(2005), and Boyd et al. (2005).
3
Our study contributes to the aforementioned literature on several dimensions. First, we
investigate the impact of important U.S. macroeconomic news on the joint distribution of the
returns in three important U.S. financial markets: equity, Treasury bonds, and corporate bonds.
This is arguably a reasonable course of action, since each of these markets does not exist in a
vacuum and investors can, and often do, hold and trade many of those securities at the same
time. Second, we use survey and futures data to extract the unexpected components of these
news announcements. Third, we analyze the impact of their release not only on the returns of the
three asset classes but also on their volatility and covariation. This effort allows us to provide a
more comprehensive picture of the effect of surprise macroeconomic information on the behavior
of asset prices in the U.S. capital markets.
We find that the process of price formation in the U.S. financial markets appears to be
driven by fundamentals, albeit often heterogeneously so. Specifically, our evidence reveals a
statistically and economically significant dichotomy, to our knowledge novel to the literature,
between the reaction of the stock and bond markets to the release of unanticipated fundamental
information. After initially falling, conditional stock return volatility increases the day this
surprise macroeconomic news is released. Conditional bond return volatility instead first rises
and then declines, the more so the shorter the maturity of the bond portfolio and the lower its
likelihood of default. We also show that conditional return comovement within and across stock
and bond markets most often decreases – rather than increasing, as commonly believed – in
correspondence with those releases, even after controlling for volatility shifts. Finally, we find that
the conditional dynamics of stock, Treasury, and corporate bond returns react asymmetrically to
the specific information content of surprise announcements, e.g., especially (but not exclusively)
to the release of “bad” news. Overall, these results shed new light on the mechanisms by which
information is incorporated into prices.
The remainder of the paper is organized as follows. Section 2 describes our database. Section
3 develops our econometric approach to estimating the impact of unexpected news releases on
asset returns. Section 4 discusses the first set of results for conditional returns, return volatility,
and return comovement. Section 5 tests for the differential effect of positive and negative news
on stock and bond return dynamics. Section 6 concludes.
2 Data
The basic dataset we use in this paper consists of a variety of daily, continuously compounded
excess holding-period returns (over three-month Treasury bills) on three asset classes – stocks,
4
Treasury bonds, and corporate bonds – whose prices are expected to be affected by four macroe-
conomic announcements: the total Consumer Price Index (CPI), the Unemployment Rate, and
the Nonfarm Payroll Employment, exogenously released on a monthly basis, and the Target
Fed Funds Rate, potentially endogenous to the former variables. Our choice of macroeconomic
announcements is motivated by several considerations. First, there is broad agreement that the
level and dynamics of employment and inflation within a country represent the main concern
of that country’s monetary authorities when setting their interest rate policy. Second, there is
ample evidence that financial markets are most sensitive to direct news about unemployment
and inflation (e.g. Krueger, 1996; Fleming and Remolona, 1999; Boyd et al. 2005; Pasquariello
and Vega, 2006). Third, the impact of other, indirect announcements about unemployment and
inflation (such as Retail Sales or Consumer Confidence) on any of the asset classes we study can
only weaken, rather than enhance the significance of our results.
We analyze seven time series of asset returns, two for the stock market, three for the Treasury
bond market, and two for the corporate bond market. Excess stock holding-period returns
(including distributions) are computed for the CRSP value-weighted portfolios made of NYSE
and AMEX stocks (rnyxt ) and of NASDAQ stocks (rnaqt ). We calculate holding-period returns
on Treasury bonds using the constant-maturity five-year, ten-year, and thirty-year time series of
interest rates constructed by the U.S. Treasury from yields on actively traded issues. These rates
are converted into excess five-year (r5yt ), ten-year (r10yt ), and thirty-year (r30yt ) holding-period
returns on hypothetical par bonds with the corresponding maturity, as in Jones et al. (1998).
A similar procedure is applied to two portfolios of corporate bonds, one made of Aaa-rated
(raaat ) and the other made of Baa-rated (rbaat ) U.S. corporate bonds, with maturity equal to or
greater than twenty years, from Moody’s Investors Service.6 Aaa-rated bonds carry the smallest
degree of investment risk. Baa-rated bonds are neither highly protected nor poorly secured,
and are generally deemed to have some speculative characteristics. Our sample covers a period
of roughly 16 years, the longest period over which data on all assets under examination are
continuously available: from January 3, 1986 (the first day for which daily Baa portfolio yields
become available) to February 14, 2002 (the last day for which constant-maturity thirty-year
Treasury bond yields are available).7 This period also largely overlaps with the tenure of Alan
Greenspan as the chairman of the Board of Governors of the Federal Reserve System.8
6For these two portfolios, we assume that the hypothetical par corporate bonds have a maturity of 20 years.7The U.S. Treasury suspended issuing thirty-year bonds in October 2001, and resumed issuing them only in
February 2006. Consequently, the thirty-year Treasury constant maturity series was discontinued on February
18, 2002 and reintroduced only on February 9, 2006.8Alan Greenspan was sworn in as chairman of the Federal Reserve in August 11, 1987, succeeding Paul Volcker.
5
Summary statistics for these series are reported in Table 1. Returns are in percentage, i.e.,
were multiplied by 100. Not surprisingly, given the growth experienced by the U.S. stock market
in the past three decades, the mean excess returns rnyxt and rnaqt are positive, significant, and the
highest among the asset classes under examination. Excess bond returns are positive as well, and
increasing with maturity and likelihood of default. Daily excess returns are also characterized by
little or no skewness, strong leptokurtosis, and small positive autocorrelation (bρ1 > 0). Finally,the Ljung-Box portmanteau test for up to the fifth-order serial correlation (LB (5)) strongly
rejects the null hypothesis that excess holding-period returns on stocks, Treasury, highest-grade
(Aaa), and medium-grade (Baa) bonds are white noise.
We assemble a database of significant macroeconomic announcements over our sample inter-
val. We collect information on all meetings of the Federal Open Market Committee (FOMC)
from Bloomberg. In particular, we focus on target Fed Funds rate decisions, which represent
the most explicit public disclosure by the Federal Reserve of its stance of monetary policy over
the sample period.9 By law, the FOMC must meet at least 4 times a year, but it has held
a minimum of 8 scheduled meetings per year since 1981. Nonetheless, Bomfim and Reinhart
(2000) observe that more than 75% of changes in the intended Fed Funds rate from 1989 to 1993
occurred between those meetings. Furthermore, until the end of 1993 the Federal Reserve made
the vast majority of its interest rate decisions in the afternoon, when the Fed Funds market in
New York was virtually closed, and declared them by conducting open-market operations the
next day. However, since March 28, 1994, the Fed Funds rate has been released regularly at 2:15
p.m. Eastern Standard Time (EST). We control for this delay by shifting the “effective” Fed
Funds announcement date by one day until then.10
Monthly real-time (actual) announcements – i.e., as reported in the original press releases
– on total CPI (cpit, in percentage in Table 2), Nonfarm Payroll changes (payt, in thousands in
Table 2), and Unemployment (unet, in percentage in Table 2) and their corresponding releases
dates are obtained from the Bureau of Labor Statistics (BLS).11 We classify these economic
9Recent studies (e.g., Boukus and Rosemberg, 2006) find that U.S. financial markets respond not only to such
quantitative information as FOMC rate decisions but also to such qualitative information as official statements,
speeches, or FOMC minutes released by the Federal Reserve to the public. This evidence, albeit consistent with
the FOMC’s increased commitment to transparency over the last few years of our sample, is sensitive to the
specific algorithm extracting common themes from multi-page, often ambiguous statements.10For more on the timing discrepancy between FOMC announcements and the end of trading in the Fed Funds
market, see Kuttner (2001) and Bernanke and Kuttner (2005).11The monthly news releases containing CPI data are usually made available to the public at 8:30 a.m. EST
of a day of the second full week of each month. News releases of monthly changes in Nonfarm Payroll and
Unemployment instead come (again at 8:30 a.m. EST) about three weeks into the month.
6
announcements according to whether their content was anticipated by the market. For that pur-
pose, we use the database of forecasts compiled by International Money Market Services (MMS).
This database is available to us for the period 1986-2002.12 Over the past two decades, MMS has
surveyed dozens of economists and money managers, collecting their forecasts for a broad range
of macroeconomic variables.13 Consistent with the existing literature on macroeconomic news
arrivals (e.g., Balduzzi et al., 2001; Andersen et al., 2003, 2005), we then define an announcement
of type e on day t, Aet , a surprise if its absolute difference with respect to the “market consen-
sus” – the corresponding median forecast F et – is “large,” i.e., greater than a predetermined
threshold. In this study, we use a threshold of 5 basis points for cpit and unet, and of 20, 000 jobs
for payt; the results that follow are not meaningfully affected by alternative thresholds since the
vast majority of expected announcements occurred at or very close to the median forecast. We
use a similar approach (and the same 5 basis points threshold) for the target rate decisions by
the Federal Reserve (fedt, in percentage in Table 2). Nonetheless, we measure the corresponding
market expectations using the thirty-day interest rate futures contract traded on the Chicago
Board of Trade (CBOT), since changes in the Fed funds futures rate – an intuitive aggregation
of market-wide policy expectations – have been shown to represent more efficient predictors of
FOMC’s target rate changes.14 We employ the algorithm devised by Kuttner (2000) to estimate
the one-day rate surprise from the one-day change in the spot-month futures rate around FOMC
announcements. Because CBOT futures prices are available only from October 3, 1988, we use
MMS forecasts for the first three years of the sample.
According to Table 2, there were 161 FOMC meetings over our sample period 1986-2002.
During that period, the Federal Reserve decided much less frequently to increase the Fed Funds
12Extending our sample beyond 2002 is problematic. Indeed, since being acquired by Informa in 2003, MMS
has stopped providing its survey services. Recently, another company, Action Economics, has resumed collecting
economists’ forecasts of U.S. macroeconomic variables, although not from the date when MMS ended its survey.
Furthermore, it is not obvious whether Action Economics has been using exactly the same survey techniques
previously employed by MMS.13MMS surveys are conducted via telephone every last Friday prior to each news announcement, and the
resulting median forecasts are released during the following week. According to several studies of their information
content (e.g., Urich and Wachtel, 1984; Balduzzi et al., 2001), MMS expectations are generally unbiased and less
noisy estimates of the corresponding realized macroeconomic variables than those generated by extrapolative
models (e.g., Figlewski and Wachtel, 1981). For a more detailed description of the MMS database and its
properties, see Andersen et al. (2003, 2005).14Krueger and Kuttner (1996), Bomfim and Reinhart (2000), and Kuttner (2001) explore the properties of
various proxies for market expectations of FOMC rate changes implied by Fed funds future rates. Piazzesi and
Swanson (2004) find that forecasts from one-day changes in Fed funds futures rates are the least affected by biases
induced by time-varying risk premia.
7
rate than to either leave it unchanged or cut it. About 44% of all rate decisions were unexpected
by the market. The CPI series suggests an annual inflation rate of 3% between 1986 and 2002.
Not surprisingly, given the behavior of the U.S. economy over the last two decades, CPI deflation
was much less common (cpit ≤ 0 only 19 times), and went almost always undetected by the
market. Recessions were of shorter length (unet > 0 in 66, and payt ≤ 0 in 40, of 190 BLS pressreleases) and generally exhibited greater per month intensity than expansions, but both were
equally difficult to predict on average (both Aunet and Apayt were sufficiently different from F unet
and F payt more than 70% of the times, regardless of their sign).
3 The empirical model
We now study the short-term anticipation and response of the U.S. stock, Treasury, and corporate
bond markets to the arrival of relevant U.S. macroeconomic news. The goal of our multi-market
analysis is to shed light not only on the effects of these announcements on conditional mean
excess returns and return volatility, but also on the comovement and interaction between those
markets in the proximity of their occurrence.
The GARCH specification proposed by Bollerslev (1986) and its many univariate and multi-
variate extensions are among the most widely adopted models to describe time-varying volatility
and covariances.15 Their success can be attributed to the ability to provide robust descriptions
of many volatility processes (see Nelson and Foster, 1994). Moreover, a large body of literature
now exists on the theoretical properties of their estimators.16 Yet, in most cases, multivariate
GARCH models are not flexible enough, and the number of parameters in them too large, to
introduce complex forms of conditional comovement among asset returns. In this paper, we use
the Dynamic Conditional Correlation (DCC) model of Engle (2002) to analyze the short-term
behavior of the U.S. financial market in proximity of the release of macroeconomic news. The
DCC specification has the flexibility of univariate GARCH models without the complexity of tra-
ditional multivariate GARCH specifications. We begin by proposing the following GARCH(1,1)
model to describe the evolution of daily excess holding-period asset returns ri:
rit = μei + ρeirit−1 + γei (0) I
et (0)S
et (0) + εit
εit =phite
it eit|zt−1 ∼ N (0, 1) (1)
hit = sit[ωei + αei
¡eit−1
¢2+ βeih
it−1],
15E.g., see Kroner and Ng (1998) for a review.16See Lumsdaine (1996) for a review.
8
where zt−1 denotes the information set at time t − 1 and sit = 1 +P+1
k=−1 δei (k) I
et (k) |Set (k)|.
According to Engle (1995, p. xii), the GARCH(1,1) is “the leading generic model for almost
all asset classes of returns.” As standard in the finance literature, Eq. (1) specifies a first-order
autocorrelation model for excess holding-period returns, to control for nonsynchronicity in prices,
microstructure effects, and gradual convergence to equilibrium. Yet, the above specification is
novel to the literature for it allows for both the arrival and the extent of macroeconomic news
surprises of type e (either fedt, cpit, unet, or payt) to affect not only conditional mean asset
returns (e.g., as in Andersen et al., 2003, 2005; Bomfim, 2003) but also the conditional variance
of excess return innovations εit.
Specifically, in Eq. (1), Iet (k) is an event dummy equal to one if a surprise macroeconomic
event of type e occurred at time t + k and equal to zero otherwise (e.g., as in Jones et al.,
1998), while Set are news surprises standardized by their sample standard deviation bσe – i.e.,
Set =Aet−F etbσe – to control for differences in units of measurement across announcements (as in
Balduzzi et al., 2001, among others), and Set (k) = Set+k if I
et (k) = 1 and zero otherwise. Thus,
the coefficient γei (0) captures the average impact of the actual surprise announcements Set (k) on
the mean excess return on the announcement dates, while the dummy coefficients δei (k) proxy
for the sequential impact of absolute standardized macroeconomic news surprises |Set (k)| onconditional return variance. The modeling choice to interact |Set (k)|, rather than Set (k), withIet (k) in the expression for h
it in Eq. (1) is motivated by existing evidence (e.g., Jones et al.,
1998; Bomfim, 2003) and preliminary analysis (available on request) indicating that the mere
occurrence of surprise macroeconomic events, regardless of their sign, affects conditional return
volatility. Hence, allowing for signed news surprises in sit would weaken both the magnitude
and significance of their estimated impact on hit.17 We can interpret the resulting coefficient
δei (1) as a measure of anticipation, i.e., as the percentage impact of the release of a unit absolute
news surprise of type e on the conditional variance of the excess return ri before that release
actually occurs at time t+1. The coefficient δei (0) is a proxy for the additional, contemporaneous
percentage impact of a unit absolute news surprise released at time t on hit. Finally, the coefficient
δei (−1) is a measure of persistence, i.e., of the additional percentage impact of the arrival of aunit absolute news surprise on hit after the information has already been revealed at time t−1.18
As mentioned earlier, we also study the impact of news arrivals on the structure of conditional
comovement among asset classes. To that purpose, measurement issues are the object of con-
17The ensuing inference for hit is unaffected by replacing Set (0) with |Set (0)| in the expression for ri in Eq. (1).
18Equivalently, we can interpret the model of Eq. (1) as the amended, conditional version of a simple uncondi-
tional event study of the behavior of the second moment of asset returns around the official dates of the release
of unexpected macroeconomic information.
9
siderable debate. In particular, Forbes and Rigobon (2002) argue that shocks to the conditional
covariance among asset returns in proximity of certain macroeconomic events may be due to
shocks to return volatility. We tackle this problem by assuming that the conditional covariance
between any two standardized residuals ηit =εit√sith
it
and ηjt =εjt√sjth
jt
at time t given zt−1, qijt , isdescribed accurately by the following exponential smoothing function:
qijt = si∗t
£λeqijt−1 + (1− λe) ηit−1η
jt−1¤, (2)
where si∗t = 1 +P+1
k=−1 dei (k) I
et (k) |Set (k)|. We add ηitη
jt to both sides of Eq. (2) and rearrange
terms to obtain an integrated moving average (IMA) process with no intercept,
ηitηjt = s
i∗t η
it−1η
jt−1 +
¡qijt − ηitη
jt
¢− λesi∗t
¡qijt−1 − ηit−1η
jt−1¢, (3)
where the errors qijt − ηitηjt , with a conditional mean of zero, are Martingale differences by con-
struction and can be thought of as white-noise sequences. Eq. (3) is an extension of the DCC
IMA model of Engle (2002) with common cross-asset adjustments (λe). More importantly, the
above model allows us to measure the anticipated (dei (1)), contemporaneous (dei (0)), and persis-
tent (dei (−1)) percentage impact of the release of a unit absolute news surprise of type e on theconditional covariance between any pair of standardized residuals ηit and ηjt in a parsimonious
way, while controlling for the impact of these arrivals on conditional volatility and mean excess
returns, as advocated by Forbes and Rigobon (2002). Indeed, any common fundamental infor-
mation shock stemming from the news arrival at time t affects directly the processes for rit and
hit in Eq. (1), via the parameters γei (0), δ
ei (−1), δei (0), and δei (1).
19
The model of Eqs. (1) and (3) provides a basic representation of the behavior of asset returns
in proximity of macroeconomic news announcements. As such, it is designed to capture only the
first-order, symmetric impact of news arrivals on the U.S. equity and bond markets. Therefore,
evidence from its estimation in support of any of the arguments discussed in Section 1 can be
interpreted as strong evidence. We investigate second-order effects, in particular the possibly
asymmetric impact of “good” and “bad” news on return dynamics, in Section 5.
As first suggested by Engle (2002), we estimate the model of Eqs. (1) and (3) in two steps.
In the first step, we estimate Eq. (1) separately for each asset i by the quasi-maximum likelihood
19In a recent paper, Andersen, Bollerslev, Diebold, and Labys (2003) argue that model-free estimates of daily
variances from intraday return data perform well in comparison with standard GARCH models of daily volatility.
Along those lines, Christiansen and Ranaldo (2005) use futures intraday data to measure realized Treasury bond-
U.S. stock return correlations in proximity of macroeconomic news announcements. However, the GARCH(1,1)-
DCC IMA model of Eqs. (1) and (3) allows us to control for the contemporaneous impact of these news arrivals
on conditional return levels when estimating realized volatility, as well as on conditional return volatility when
estimating realized return correlations.
10
(QML) procedure described in Bollerslev and Wooldridge (1992). Hence, the resulting estimates
are consistent and asymptotically efficient. In the second step, we estimate the parametric
model of Eq. (3) jointly for all assets i, again by quasi-maximum likelihood, using the parameters
obtained in the first step. Under some mild regularity conditions, consistency of those parameters
ensures consistency, although not efficiency, of the estimates stemming from the second step.
Engle and Sheppard (2001), Engle (2002), and Cappiello et al. (2004) show that these estimates
perform well in a variety of situations and provide reasonable empirical results with minimal loss
of efficiency.20
4 The basic results
We start our analysis by estimating the model of Eqs. (1) and (3) for each of the four macroe-
conomic announcements in our sample according to the two-step procedure described in Section
3. We report the resulting QML estimates (and robust t-statistics) of selected parameters of
Eqs. (1) and (3) in Tables 3 and 4, respectively, when e = fed, cpi, une, or pay. The inclusion
of dummies to control for day-of-the-week effects or the day-level clustering of other macroeco-
nomic announcements (e.g., Jones et al., 1998; Kim et al., 2004) does not significantly affect our
inference, either qualitatively or quantitatively.21
Table 3 reveals a dichotomy between the reaction of the U.S. stock and bond markets to
economic news. Conditional stock return volatility is somewhat lower the day before (δei (1) < 0),
and significantly higher the day of their arrival (δei (0) > 0). In particular, our estimates indicate
that between 1986 and 2002, upon the release of a unit absolute standardized macroeconomic
news surprise (with the exception of e = fed), conditional return volatility for the NYSE-AMEX
(NASDAQ) is on average no less than 50% (35%) and almost 100% (60%) higher than during non-
20Efficiency could be achieved by estimating jointly the entire set of moment conditions from Eqs. (1) and (3).
Yet, convergence is often problematic since the resulting log-likelihood function is often “extremely flat” (Cajigas
and Urga, 2006). Joint estimation of our basic model and its extensions (when successful, in unreported analysis)
leads to virtually identical inference.21We do not report these additional analyses for brevity; yet, they are available on request from the authors.
Furthermore, the timing of the macroeconomic releases in our sample is not clustered and (with the exception of
the few unscheduled FOMC meetings over our sample period) can be deemed exogenous to financial markets. Yet,
our model does not allow to assess whether the clustering of various unexpected macroeconomic news releases
in the same direction, over a short period of time, may have larger effects on U.S. financial markets than the
average effect of each news release separately, as stemming from the estimation of Eqs. (1) and (3). In these
circumstances, our approach is likely to bias such effects downward. We thank an anonymous referee for pointing
out this possibility to us.
11
announcement days. This is significant since those surprise announcement windows are frequent,
accounting for about 21% of all days in our sample (see Tables 1 and 2). In contrast, we find
that the coefficients δei (1) are always positive and mostly statistically significantly for conditional
government and corporate bond return volatility. For instance, the day before the release of a
unit absolute unemployment news (e = une or pay), return volatility increases by a minimum
of 62% (δpayi (1) = 0.621) for Aaa-rated long-term corporate bonds to a maximum of 115%
(δpayi (1) = 1.152) for five-year Treasury bonds. This suggests that there is considerable increase
in uncertainty among bond market participants in anticipation of the release of macroeconomic
news. This effect is only partially short-lived, since the contemporaneous coefficients δei (0), albeit
often significantly negative (at the 1% level), are of much lower magnitude (e.g., δpayi (0) = −0.249for five-year Treasury bonds), while δei (−1) is either small, negative, and weakly significant, orzero. Stock return volatility instead declines appreciably after the announcements, especially for
NYSE-AMEX (δcpii (−1) = −0.189, δunei (−1) = −0.119, and δpayi (−1) = −0.221 in Table 3).This evidence indicates that only in the U.S. bond market any additional information stemming
from economic news does not appear to augment price fluctuations, i.e., that this news appears to
induce (at least partial) resolution of uncertainty and/or disagreement only among bond market
participants.
The arrival of employment news has the greatest impact on the government bond market,
especially before their release, while CPI news has the lowest; the corresponding effect on corpo-
rate bonds, although of same sign, is of somewhat smaller scale. Interestingly, Nonfarm Payroll
surprises, although released simultaneously with unemployment numbers, are preceded and ac-
companied by relatively more pronounced volatility shocks. This evidence is consistent with
some recent studies (e.g., Andersen et al., 2005; Pasquariello and Vega, 2007) suggesting that
Nonfarm Payroll has the greatest information content among all public signals of the state of the
U.S. economy available to investors and speculators. The milder reaction of both government
and corporate bonds to CPI surprises may instead stem from the relative stability of inflation
expectations and the Fed’s significant credibility in fighting inflation over our sample period 1986-
2002. Yet, the evidence in Table 3 also suggests that the behavior of stock and bond markets
close to target rate decisions by the Federal Reserve is both less economically and statistically
significant than for any other event in our sample.
The absolute magnitude of δei (1) and δei (0) is decreasing in the maturity of the bond portfolios
and in the likelihood of default, as proxied by the Moody’s ratings. The latter is somewhat
surprising, given the greater sensitivity of corporate bonds of poorer quality to the business
cycle (e.g., Gertler and Lown, 2000). The notion that five-year Treasury bonds are the most
12
liquid (e.g., Fleming, 2003; Brandt and Kavajecz, 2004) may instead explain the former, since
we would expect the intensity of information gathering and portfolio rebalancing induced by the
news arrivals to be greater for more liquid securities.22 Accordingly, target rate decisions have
a positive and statistically significant impact on the conditional return volatility exclusively of
five-year Treasury bonds (δfedi (0) = 0.450). Alternatively, mean reversion in short-term interest
rates (e.g., Chapman and Pearson, 2000 and references therein) may lead to a weaker impact
of an information shock on longer-term bonds, since any impact of such shocks on the short
end of the yield curve is expected to die out for its long end. U.S. stock markets instead
display a remarkable homogeneity of responses to most surprise macroeconomic announcements.
Specifically, the impact of their arrival on NYSE-AMEX conditional return and volatility is
qualitatively and quantitatively similar to that estimated for NASDAQ, despite the fact that
most “new economy” companies are usually deemed less sensitive to employment or inflation
news than to the state of their specific industry. For instance, the conditional return volatility
for both NYSE-AMEX and NASDAQ decline by roughly 13% on average (δfedi (−1) = −0.134for rnyxt and δfedi (−1) = −0.127 for rnaqt in Table 3) following unit absolute surprise Fed rate
decisions.
At this stage of the analysis, we find weak or no evidence of a relation between the condi-
tional mean excess holding-period returns of each of the asset classes we study and the release
of macroeconomic news surprises. In particular, Table 3 shows that the corresponding estimated
coefficient γi (0) is insignificant for all but (NASDAQ) stocks (γfedi (0) = −0.135 for rnaqt ), as
in Bomfim (2003). Table 4 provides evidence of a much greater dichotomy between the stock
and the bond markets when analyzing our estimates for the parameters of the DCC IMA co-
variance model of Eq. (3). There we report intra and across-asset class averages of both QML
estimates of the dummy coefficients dei (k) in Eq. (3) and their robust t-statistics. These av-
erage estimates suggest that the release of the surprise economic announcements in our sample
is preceded by sharply lower comovement among U.S. stock markets (e.g., dunei (1) = −1.267),but by much less so among government and corporate bond markets. Declining comovement
among NYSE-AMEX and NASDAQ stocks is often magnified on the day of news arrivals and
the day afterward (e.g., dunei (0) = −1.138 and dunei (−1) = −0.206), even after controlling fortheir impact on conditional stock return and return volatility. It is however during those days
that comovement among government bond returns and among corporate bond returns drops most
22E.g., Chowdhry and Nanda (1991). Yet, shorter-maturity bonds are also less sensitive to fluctuations of
interest and inflation rate expectations potentially induced by that news. According to Table 3, this effect is
weaker than the maturity (or liquidity) effect.
13
significantly for all announcement types. The dynamics of intra-asset class comovement of ex-
cess holding-period returns are relatively homogeneous in intensity and significance except when
e = cpi. For instance, it is only in those circumstances that the estimated pre-announcement
decoupling among NYSE-AMEX and NASDAQ returns is partially reversed (dcpii (1) = −0.083but dcpii (0) = 0.564 and dcpii (−1) = 0.700), while the estimated announcement-day decline in
comovement among government bond returns is the most persistent (dcpii (−1) = −2.685) andthe weakest among Moody’s portfolios of Aaa and Baa-rated corporate bonds (dcpii (0) = −0.255and dcpii (−1) = −0.106).Finally, comovement among all asset classes decreases in the proximity of all macroeconomic
news releases. For example, we find that comovement between stock and government bonds,
stocks and corporate bonds, and government and corporate bonds declines on average by 103%,
93%, and 150%, respectively, in correspondence with the release of a unit absolute standardized
unemployment news surprise (dunei (0) = −1.033, dunei (0) = −0.934, and dunei (0) = −1.504 inTable 4). CPI and Nonfarm Payroll surprises lead to negative comovement shocks of similar
magnitude. Fed rate decisions are instead accompanied by the smallest (yet still economically
significant) drop in across-asset class covariance, consistent with the results reported in Table 3.
This evidence is important for it suggests that any multi-asset trading activity due to portfolio
rebalancing, information spillover, financial constraints, or shifts to the degree of information
asymmetry and heterogeneity among traders induced by that news translates into either more
negative or less positive – but not more positive, as commonly thought – comovement among
the excess holding-period returns in our sample.
5 Asymmetric news impact
In the previous section, we showed that the arrival and absolute magnitude of surprise macroe-
conomic announcements affected significantly the U.S. equity and bond markets between 1986
and 2002. The sign and magnitude of the impact of their release on the dynamics of asset prices
should also depend on their specific information content. Indeed, it is reasonable to conjecture
that the effect of “good” or “bad” news on the process of price formation in the U.S. financial
markets may be asymmetric.23 For example, Boyd et al. (2005) find that, in the short run, the
reaction of stock and bond prices to unemployment news, measured with respect to a statistical
model for final release numbers, is state-dependent. The equity market usually responds posi-
23In light of earlier discussion, we use the terms “good” and “bad” to refer exclusively to the content of the
news, rather than to its implications for stock and bond valuations.
14
tively to surprisingly rising unemployment, while Treasury bond prices react to it only during
expansions. Similarly, Veronesi (1999) argues that “good” or “bad” news may increase uncer-
tainty when released in “bad” or “good” times, respectively.24 Furthermore, it is also possible
that the various forms of market frictions and financial constraints described in the literature
(e.g., borrowing, short-selling, and wealth constraints) are more binding on investors’ behavior
following “negative” announcements. These constraints could then affect not only the level of
asset returns (e.g., Diether et al., 2002) but also their volatility and comovement (e.g., Kyle and
Xiong, 2001) differentially in proximity of news arrivals.
We explore these issues by amending the model of Eqs. (1) and (3) to allow for asymmetric ef-
fects of news releases. Specifically, we introduce the following GARCH(1,1) model for conditional
return volatility
rit = μi + ρirit−1 + [γ
ei (0,+) I
et (0,+) + γei (0,−) Iet (0,−)]Set + εit
εit =
qsi±t e
it eit|zt−1 ∼ N
¡0, hit
¢(4)
hit = ωi + αi¡eit−1
¢2+ βih
it−1,
where si±t = 1 +P+1
k=−1 δei (k,+) I
et (k,+) |Set (k,+)| +
P+1k=−1 δ
ei (k,−) Iet (k,−) |Set (k,−)|, and
then model comovement among excess return residuals as
ηitηjt = s
i±∗t ηit−1η
jt−1 +
¡qijt − ηitη
jt
¢− λesi±∗t
¡qijt−1 − ηit−1η
jt−1¢, (5)
where si±∗t = 1 +P+1
k=−1 dei (k,+) I
et (k,+) |Set (k,+)|+
P+1k=−1 d
ei (k,−) Iet (k,−) |Set (k,−)|, ηit =
εit√si±t hit
, Iet (k,+) and Set (k,+) are dummy variables equal to one and S
et+k, respectively, if the
Federal Reserve announced either a surprisingly large rate increase or a surprisingly small rate
cut on day t + k (Sfedt+k > 0), or if the CPI or the unemployment rate was reported at day t + k
to have either increased surprisingly highly or decreased surprisingly little with respect to the
previous month (Scpit+k > 0 or Sunet+k > 0), or if the Nonfarm Payroll was reported on day t + k
to have either decreased surprisingly highly or increased surprisingly little with respect to the
previous month (Spayt+k < 0), and zero otherwise; vice versa, Iet (k,−) and Set (k,−) are dummy
variables equal to one and Set , respectively, if the Federal Reserve announced either a surprisingly
large rate cut or a surprisingly small rate increase on day t+ k (Sfedt+k < 0), or if the CPI or the
unemployment rate was reported at day t + k to have either decreased surprisingly highly or
increased surprisingly little with respect to the previous month (Scpit+k < 0 or Sunet+k < 0), or if
the Nonfarm Payroll was reported on day t + k to have either increased surprisingly highly or
24Further, Andersen et al. (2005) report that, during the 1990s, intraday stock, bond, and currency futures
returns experienced heterogeneous conditional mean and volatility jumps in presence of “good” or “bad” news.
15
decreased surprisingly little with respect to the previous month (Spayt+k > 0), and zero otherwise.
Hence, Sfedt (k,+) (Ifedt (k,+)), Scpit (k,+) (Icpit (k,+)), Sunet (k,+) (Iunet (k,+)), and Spayt (k,−)(Ipayt (k,−)) are “bad” news (dummies), while Sfedt (k,−) (Ifedt (k,−)), Scpit (k,−) (Icpit (k,−)),Sunet (k,−) (Iunet (k,−)), and Spayt (k,+) (Ipayt (k,+)) are “good” news (dummies).25 According
to Table 2, “bad” news so defined is generally less frequent but of greater absolute magnitude
than “good” news over our sample period 1986-2002.
We estimate the model of Eqs. (4) and (5) according to the two-stage QML procedure
described in Section 3 and report the ensuing coefficients in Table 5 for conditional returns
and volatility and in Table 6 for return covariances. Our results reveal a significant degree of
asymmetry in the response of the U.S. financial markets to the arrival of macroeconomic news
of positive versus negative information content. Table 5 shows that the absolute magnitude of
the effects of news releases on return volatility described in Section 4 is generally, although not
homogeneously, greater when the news is “bad” (especially “bad” CPI and Nonfarm Payroll
news: Scpit (k,+) > 0 and Spayt (k,−) < 0): Conditional stock (bond) return volatility rises
more sharply the day of (before), and drops more sharply the day after (of) their release. For
instance, we find that the conditional volatility of thirty-year Treasury bond returns increases
by 77% (δpayi (1,−) = 0.766) the day before, and then declines by 22% (δpayi (0,−) = −0.215)the day of the release of surprisingly “bad” Nonfarm Payroll numbers (Spayt (k,−) < 0), versusan increase of 59% (δpayi (1,+) = 0.589) and subsequent decline by 21% (δpayi (0,+) = −0.205)in correspondence with unexpectedly “good” Nonfarm Payroll announcements (Spayt (k,+) > 0).
Target rate decisions by the Federal Reserve represent a noteworthy exception, for their absolute
impact on the dynamics of conditional stock and bond return volatility is greater when more
expansionary (or less tightening) than expected (Sfedt (k,−) < 0). Overall, our evidence suggeststhat negative macroeconomic news is more likely to induce greater uncertainty in the U.S. stock
and bond markets.
According to Table 3, neither the occurrence nor the absolute magnitude of important macroe-
conomic news affect conditional mean excess holding-period returns for U.S. stocks and bonds
between 1986 and 2002. In contrast, Table 5 provides evidence of significant asymmetric effects
of signed news surprises on those returns. In particular, the estimates reported in Table 5 suggest
that most surprisingly “good” macroeconomic announcements significantly increase conditional
mean stock returns, but decrease conditional mean Treasury and corporate bond returns. Yet,
25As in Section 3, we interact |Set (k,±)| with Iet (k,±) in the expressions for hit and ηitηjt in Eqs. (4) and (5),
respectively, to allow for a meaningful comparison of the resulting estimates for δei (k,±) and dei (k,±) with thecorresponding dummy coefficients δei (k) and d
ei (k) from Eqs. (1) and (3) reported in Tables 3 and 4.
16
most unexpectedly “bad” announcements have no meaningful contemporaneous impact on any
rit. For example, we find that estimates for the contemporaneous impact of unexpectedly expan-
sionary rate decisions by the Federal Reserve (Sfedt (k,−) < 0) on mean excess stock returns arenegative and both statistically and economically significant (−19 and −25 basis points, respec-tively, i.e., γfedi (0,−) = −0.190 for rnyxt and γfedi (0,−) = −0.246 for rnaqt in Table 5). Better
than expected inflation news (Scpit (k,−) < 0) – that generally accompanies a slowing economy
and may lead to future target rate cuts – has no impact on mean excess stock returns but lead
to a small decrease in conditional five-year Treasury and Baa-rated corporate bond returns (by
4 and 6 basis points, respectively, i.e., γcpii (0,−) = 0.044 for r5yt and γcpii (0,−) = −0.058 forrbaat in Table 5). Accordingly, worse than expected Nonfarm Payroll numbers (Spayt (k,−) < 0)also lead to lower mean excess government bond returns. However, “good” employment news
(Sunet (k,−) < 0 and Spayt (k,+) > 0) – that usually accompanies a growing economy and may
lead to future target rate increases – significantly decreases only mean excess NASDAQ and
NYSE-AMEX returns, respectively (by 17 and 19 basis points, i.e., γunei (0,+) = −0.172 for rnaqt
and γpayi (0,+) = −0.194 for rnyxt in Table 5).
Finally, the differential information content of macroeconomic news has little or no impact
on the (negative) direction of the shocks to comovement across asset classes induced by their
arrival we discussed in Section 4. However, the estimates of the DCC IMA model of Eq. (5)
in Table 6 reveal a significant heterogeneity in the intensity of the impact of the release of
“good” or “bad” news on intra and across-asset class covariances, especially in proximity of the
release of CPI data. For example, the comovement between government bond holding-period
returns declines by 66% on the day following better than expected inflation news Scpit (k,−) < 0(dcpii (−1,−) = −0.656) – in line with estimates for other news releases in our sample – but
by much more (dcpii (−1,+) = −3.375) immediately after worse than expected inflation newsScpit (k,+) > 0. Similarly, although lower comovement among NYSE-AMEX and NASDAQ stock
returns precedes surprisingly expansionary Fed Funds rate decisions as well (Sfedt (k,−) < 0),
the estimated decline in intra-asset class return covariances is again more pronounced and per-
sistent in proximity of a tighter than expected monetary policy stance by the Federal Reserve
(Sfedt (k,+) > 0). Comovement within and across all asset classes also displays more pronounced
downward dynamics when news about worsening employment conditions (Sunet (k,+) > 0 and
Spayt (k,−) < 0) is released. This suggests that surprisingly “bad” news arrivals are more likelyto induce a wave of rebalancing activity among the asset classes in our sample. Such asymmetry
is consistent with recent studies (e.g., Kyle and Xiong, 2001) conjecturing that the multi-asset
trading activity of informed and uninformed market participants is more likely to respond to
17
financial constraints in those circumstances. Accordingly, target rate increases and “bad” Non-
farm Payroll news induce lower comovement among high-rated and low-rated corporate bonds
than rate cuts and “good” employment numbers, i.e., are more likely to induce investors to shift
their portfolios away from low-quality issuers.
6 Conclusions
The analysis of the extent to which prices in financial markets incorporate fundamental informa-
tion is central to the theoretical and empirical finance literature. The traditional notion of market
efficiency requires that new information about asset payoffs should be quickly and fully reflected
in asset prices and drive their dynamics. Prior research has examined the links between financial
and real variables by studying the effects of the disclosure of macroeconomic information (often
without first identifying its surprise content) on stock and bond markets (often separately).
Our paper contributes to this debate by providing a comprehensive analysis of the impact of
the unexpected component of important U.S. macroeconomic news releases on the process of price
formation in the three most important U.S. financial markets, for equity, government bonds, and
corporate bonds. For that purpose, we develop an extension of the GARCH-DCC IMA model
of Engle (2002) that allows us to simultaneously (yet parsimoniously) identify the effects of
news arrivals on conditional mean returns, return volatility, and return covariance. We estimate
different versions of this model over our sample of announcements of target rate decisions by
the Federal Reserve, CPI, Unemployment, and Nonfarm Payroll data from the Bureau of Labor
Statistics between 1986 and 2002.
We find that the arrival of surprise macroeconomic news has a statistically and economically
significant impact on the U.S. financial markets, but also that this impact varies greatly across
asset classes. Conditional stock return volatility decreases on the trading day before, increases
on the day when the announcements are made, and subsequently decreases. Conditional bond
return volatility instead increases when the news is released, and declines afterward. This effect
is stronger for portfolios of bonds of shorter maturity (typically, the most liquid and sensitive to
the mean-reverting nature of the short rate drift) or less likely to default. The estimated shifts
in volatility appear to be persistent in the short run, i.e., do not offset each other completely
over a three-day event window around the announcements. These effects are also asymmetric:
Their absolute magnitude is generally greater when the macroeconomic information released
represents “bad” news. Conditional mean excess holding-period returns for stock and bonds are
instead mostly sensitive to the release of unexpectedly “good” news. Finally, our estimates paint
18
a complex picture of the interaction between asset returns in proximity of macroeconomic news
releases. Yet, they offer little or no support to the commonly held notion that the arrival of this
news is accompanied by greater comovement among asset returns. Indeed, return comovement
within and across stock and bond markets most often decreases in correspondence with those
announcements, especially in proximity of “bad” news.
Ultimately, this study reports a wide array of novel empirical evidence – stemming from a
robust yet manageable methodology– on the effects of the release of macroeconomic news on the
moments of returns in U.S. equity, government bond, and corporate bond markets. We suggest
that some of this evidence is consistent with existing theoretical arguments in the literature, e.g.,
those emphasizing the role of investors’ trading activity. Nonetheless, we hope that our work
may stimulate further research on a unifying theory to explain all of these fascinating stylized
facts.
References
Andersen, T., Bollerslev, T., Diebold, F., Labys, P. (2003). Modeling and Forecasting Realized
Volatility. Econometrica, 71, 579-625.
Andersen, T., Bollerslev, T., Diebold, F., Vega, C. (2003). Micro Effects of Macro Announce-
ments: Real-Time Price Discovery in Foreign Exchange. American Economic Review, 93, 1132-
1145.
Andersen, T., Bollerslev, T., Diebold, F., Vega, C. (2005). Real Price Discovery in Stock, Bond,
and Foreign Exchange Markets. Journal of International Economics, forthcoming.
Andersen, T., Bollerslev, T., Meddahi, N. (2005). Correcting The Error: Volatility Forecast
Evaluation Using High-Frequency Data and Realized Volatilities. Econometrica, 73, 279-296.
Bai, X., Russell, J., Tiao, G. (2000). Beyond Merton’s Utopia: Effects of Non-Normality and De-
pendence on the Precision of Variance Estimates Using High-Frequency Financial Data. Working
Paper, University of Chicago.
Balduzzi, P., Elton, E., Green, C. (2001). Economic News and Bond Prices: Evidence from the
U.S. Treasury Market. Journal of Financial and Quantitative Analysis, 36, 523-543.
Bernanke, B., Kuttner, K. (2005). What Explains the Stock Market’s Reaction to Federal Reserve
Policy? Journal of Finance, 60, 1221-1257.
19
Bollerslev, T. (1986). Generalized Autoregressive Conditional Heteroskedasticity. Journal of
Econometrics, 31, 307-327.
Bollerslev, T., Wooldridge, J. (1992). Quasi-Maximum Likelihood Estimation and Inference in
Dynamic Models with Time-Varying Covariances. Econometric Reviews, 11, 143-172.
Bomfim, A. (2003). Pre-announcement Effects, News Effects, and Volatility: Monetary Policy
and the Stock Market. Journal of Banking & Finance, 27, 133-151.
Bomfim, A., Reinhart, V. (2000). Making News: Financial Market Effects of Federal Reserve
Disclosure Practices. Working Paper, IMF.
Boukus, E., Rosemberg, J. (2006). The Information Content of FOMC Minutes. Working Paper,
Federal Reserve Bank of New York.
Boyd, J., Hu, J., Jagannathan, R. (2005). The Stock Market’s Reaction to Unemployment News:
Why Bad News is Usually Good for Stocks? Journal of Finance, 60, 649-672.
Brandt, M., Kavajecz, K. (2004). Price Discovery in the U.S. Treasury Market: The Impact of
Order Flow and Liquidity on the Yield Curve. Journal of Finance, 59, 2623-2654.
Cajigas, J., Urga, G. (2006). Dynamic Conditional Correlation Models with Asymmetric Multi-
variate Laplace Innovations. Working Paper, Cass Business School.
Cappiello, L., Engle, R. Sheppard, K. (2004). Asymmetric Dynamics in the Correlations of Global
Equity and Bond Returns. ECB Working Paper No. 204.
Chapman, D., Pearson, N. (2000). Is the Short Rate Drift Actually Nonlinear? Journal of
Finance, 55, 355-388.
Chen, N., Roll, R., Ross, S. (1986). Economic Forces and the Stock Market. Journal of Business,
59, 383-403.
Chowdhry, B., Nanda, V. (1991). Multimarket Trading andMarket Liquidity.Review of Financial
Studies, 4, 483-511.
Christiansen, C., Ranaldo, A. (2005). Realized Bond-Stock Correlation: Macroeconomic An-
nouncement Effects. Working Paper, Swiss National Bank.
Connolly, R., Stivers, C., Sun, L. (2005). Stock Market Uncertainty and the Stock-Bond Return
Relation. Journal of Financial and Quantitative Analysis, 40, 161-194.
20
Connolly, R., Stivers, C., Sun, L. (2007). Commonality in the Time Variation of Stock-Stock and
Stock-Bond Return Comovements. Journal of Financial Markets, 10, 192-218.
Connolly, R., Wang, A. (2000). On Stock Market Return Co-movements: Macroeconomic News,
Dispersion of Beliefs, and Contagion. Working Paper, Kenan-Flagler Business School.
Diether, K., Malloy, C., Scherbina, A. (2002). Differences of Opinion and the Cross Section of
Stock Returns. Journal of Finance, 57, 2113-2141.
Engle, R. (1995). ARCH: Selected Readings. Oxford University Press.
Engle, R. (2002). Dynamic Conditional Correlation - A Simple Class of Multivariate GARCH
Models. Journal of Business and Economic Statistics, 20, 339-350.
Engle, R., Sheppard, K. (2001). Theoretical and Empirical Properties of Dynamic Conditional
Correlation Multivariate GARCH. NBER Working Paper No. 8554.
Fama, E. (1971). Risk, Return, and Equilibrium. Journal of Political Economy, 79, 30-55.
Figlewski, S., Wachtel, P. (1981). The Formation of Inflationary Expectations. Review of Eco-
nomics and Statistics, 63, 1-10.
Flannery, M., Protopapadakis, A. (2002). Macroeconomic Factors Do Influence Aggregate Stock
Returns. Review of Financial Studies, 15, 751-782.
Fleming, J., Kirby, C., Ostdiek, B. (1998). Information and Volatility Linkages in the Stock,
Bond, and Money Markets. Journal of Financial Economics, 49, 111-137.
Fleming, M. (2003). Measuring Treasury Market Liquidity. Economic Policy Review, Federal
Reserve Bank of New York, 9, 83-108.
Fleming, M., Piazzesi, M. (2005). Monetary Policy Tick-by-Tick. Working Paper, Federal Reserve
Bank of New York.
Fleming, M., Remolona, E. (1999). Price Formation and Liquidity in the U.S. Treasury Market:
The Response to Public Information. Journal of Finance, 54, 1901-1915.
Forbes, K., Rigobon, R. (2002). No Contagion, Only Interdependence: Measuring Stock Market
Co-Movements. Journal of Finance, 57, 2223-2261.
Foster, F., Viswanathan, S. (1993). The Effect of Public Information and Competition on Trading
Volume and Price Volatility. Review of Financial Studies, 6, 23-56.
21
Gertler, M., Lown, C. (2000). The Information in the High Yield Bond Spread for the Business
Cycle: Evidence and Some Implications. NBER Working Paper No. 7549.
Green, C. (2004). Economic News and the Impact of Trading on Bond Prices. Journal of Finance,
59, 1201-1233.
Hasbrouck, J. (2006). Empirical Market Microstructure: The Institutions, Economics and Econo-
metrics of Securities Trading. Oxford University Press.
Jones, C., Lamont, O., Lumsdaine, R. (1998). Macroeconomic News and Bond Market Volatility.
Journal of Financial Economics, 47, 315-337.
Kallberg, J., Pasquariello, P. (2007). Time-Series and Cross-Sectional Excess Comovement in
Stock Indexes. Journal of Empirical Finance, forthcoming.
Karolyi, G., Stulz, R. (1996). Why Do Markets Move Together? An Investigation of U.S.-Japan
Stock Return Comovements. Journal of Finance, 51, 951-986.
Kim, S., McKenzie, M., Faff, R. (2004). Macroeconomic News Announcements and the Role of
Expectations: Evidence for US Bond, Stock and Foreign Exchange Markets. Journal of Multi-
national Financial Management, 14, 217-232.
King, M., Wadhwani, S. (1990). Transmission of Volatility between Stock Markets. Review of
Financial Studies, 3, 5-33.
Kodres, L., Pritsker, M. (2002). A Rational Expectations Model of Financial Contagion. Journal
of Finance, 57, 769-799.
Kroner, K., Ng, V. (1998). Modeling Asymmetric Comovements of Asset returns. Review of
Financial Studies, 11, 817-844.
Krueger, A., (1996). Do Markets Respond More to More Reliable Labor Market Data? A Test
of Market Rationality. NBER Working Paper No. 5769.
Krueger, J., Kuttner, K. (1996). The Fed Funds Futures Rate as a Predictor of Federal Reserve
Policy. Journal of Futures Markets, 16, 865-879.
Kuttner, K. (2001). Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds
Futures Market. Journal of Monetary Economics, 47, 523-544.
Kyle, A., Xiong, W. (2001). Contagion as a Wealth Effect. Journal of Finance, 56, 1401-1440.
22
Li, L., Engle, R. (1998). Macroeconomic Announcements and Volatility of Treasury Futures.
Working Paper, University of California, San Diego.
Merton, R. (1973). An Intertemporal Capital Asset Pricing Model. Econometrica, 41, 867-887.
Nelson, D., Foster, D. (1994). Asymptotic Filtering Theory for Univariate ARCHModels. Econo-
metrica, 62, 1-41.
Pasquariello, P. (2007). Imperfect Competition, Information Heterogeneity, and Financial Con-
tagion. Review of Financial Studies, 20, 391-426.
Pasquariello, P., Vega, C. (2007). Informed and Strategic Order Flow in the Bond Markets.
Review of Financial Studies, 20, 1975-2019.
Piazzesi, M., Swanson, E. (2004). Futures Prices as Risk-Adjusted Forecasts of Monetary Policy.
NBER Working Paper No. 10547.
Pindyck, R., Rotemberg, J. (1990). The Excess Comovement of Commodity Prices. Economic
Journal, 100, 1173-1189.
Pindyck, R., Rotemberg, J. (1993). The Comovement of Stock Prices. Quarterly Journal of
Economics, 108, 1073-1104.
Ross, S. (1989). Information and Volatility: The No-Arbitrage Martingale Approach to Timing
and Resolution Irrelevancy. Journal of Finance, 44, 1-17.
Shiller, R. (1989). Comovements in Stock Prices and Comovements in Dividends. Journal of
Finance, 44, 719-729.
Urich, T., Wachtel, P. (1984). The Effects of Information and Money Supply Announcements on
Interest Rates. Journal of Finance, 39, 1177-1188.
Vega, C. (2006). Stock Price Reaction to Public and Private Information. Journal of Financial
Economics, 82, 103-133.
Veronesi, P. (1999). Stock Market Overreaction to Bad News in Good Times: A Rational Ex-
pectations Equilibrium Model. Review of Financial Studies, 12, 975-1007.
23
Table1.Descriptivestatistics:Excessreturns
Thistablereportssummarystatisticsforthetimeseriesofdailycontinuouslycompoundedexcessholding-periodreturnsforthevalue-weighted
portfoliosofNYSE
andAMEXstocks( rnyx
t),NASDAQstocks,(rnyx
t),constantmaturity5yearTreasurybonds(r5yt),constantmaturity10year
Treasurybonds( r10y
t),constant-maturity30yearTreasurybonds(r30y
t),Aaa-ratedcorporatebonds(raaa
t),andBaa-ratedcorporatebonds(rbaa
t),
describedinSection3,overthethree-monthTreasurybill,forthesamplefrom
01/03/1986to02/14/2002(thelastdayforwhichr30y
tisavailable).
Returnsareexpressedinpercentage,i.e.,multipliedby100.Nisthenumberofavailableobservationsforeachseries.μisthemeanandσis
thestandarddeviation.MinandMax
aretheminimum
andmaximum
valuefortheseriesoverthesample.Skew
isthecoefficientofskewness,
whileKurtistheexcesskurtosis;theirstandarderrorsforasymptoticnormaldistributionsare¡ 6 N¢1 2 an
d¡ 24 N¢1 2 ,re
spectively.bρ 1is
theestimated
first-orderautocorrelation,whileLB(5)isthevalueoftheLjung-Boxtestofrandomnessforuptothefifth-orderautocorrelation,asymptotically
distributedas
χ2[5]underthenullhypothesisthattheseriesiswhitenoise.A“∗”,“∗∗”,or“∗∗∗”indicatesignificanceatthe10%,5%,or1%
level,respectively.
Dailyexcessholding-periodreturns
rnyx
trnaq
tr5yt
r10y
tr30y
traaa
trbaa
t
μ0.053∗∗∗
0.056∗∗
0.030∗∗∗
0.034∗∗∗
0.039∗∗
0.040∗∗∗
0.043∗∗∗
σ0.940
1.442
0.280
0.451
0.685
0.424
0.393
Min
-18.233
-11.453
-1.721
-2.774
-3.882
-2.991
-3.039
Max
8.799
14.234
2.948
4.629
7.191
3.405
1.973
bρ 10.085∗∗∗
0.082∗∗∗
0.063∗∗∗
0.054∗∗∗
0.036∗∗
0.073∗∗∗
0.059∗∗∗
LB(5)
45.37∗∗∗
41.67∗∗∗
24.77∗∗∗
25.59∗∗∗
15.10∗∗∗
37.37∗∗∗
25.18∗∗∗
Skew
-2.181∗∗∗
-0.123∗∗∗
0.132∗∗∗
-0.031
0.156∗∗∗
-0.304∗∗∗
-0.511∗∗∗
Kurt
41.565∗∗∗
9.952∗∗∗
5.427∗∗∗
4.908∗∗∗
5.147∗∗∗
4.846∗∗∗
3.803∗∗∗
N4069
4069
4069
4069
4069
4069
4069
24
Table2.Descriptivestatistics:Macroeconomicevents
Thistablereportssummarystatisticsforthefollowingmonthlypreliminarynewsreleases:FederalReservetargetrates(fed,inpercentage),
targetrateincreases( fed+),targetratedecreasesorunchanged(fed−),ConsumerPriceIndex(CPI)inflation(cpi,inpercentage)from
theBureau
ofLaborStatistics(BLS),CPIincreases( cpi+),CPIdecreasesorunchanged(cpi−),unemploymentratechange(une,inpercentage)from
theBLS,
unemploymentrateincrease( une+),decreasedorunchangedunemploymentrate(une−),NonfarmPayrollEmploymentchange(pay,inthousands)
from
theBLS,increaseinNonfarm
PayrollEmployment( pay+),anddecreasedorunchangedNonfarm
PayrollEmployment(pay−).Thesample
isfrom
01/03/1986to02/14/2002.Nisthenumberofavailableobservationsforeachseries.μisthemean,Med
isthemedian,andσisthe
standarddeviationoftheannouncedvariable;MinandMax
aretheminimum
andmaximum
valuefortheannouncedvariableoverthesample.
Marketexpectationsfortheseannouncementsarefrom
MMS;announcementsurprisesarecomputedwithrespecttomarketconsensusaccordingto
aproceduredescribedinSection3.
μsur,Medsur,σsur,andNsurarethemean,median,standarddeviation,andnumberofavailableobservations
ofthesurprisedifference(whenpositive,+,negative,−,orboth)betweentheannouncedandexpectedamount.A“∗”,“∗∗”,or“∗∗∗”indicate
significanceatthe10%,5%,or1%
level,respectively.
Macroeconomicevents
fed
fed+
fed−
cpi
cpi+
cpi−
une
une+
une−
pay
pay+
pay−
μ-0.034∗
0.342∗∗∗
-0.117∗∗∗
0.256∗∗∗
0.294∗∗∗
-0.084∗∗∗
-0.006
0.179∗∗∗
-0.105∗∗∗
146.1∗∗∗
216.8∗∗∗
-118.9∗∗∗
Med
0.000
0.250
0.000
0.200
0.300
0.000
0.000
0.100
-0.100
167.5
209.0
-96.00
σ0.246
0.149
0.175
0.201
0.168
0.146
0.170
0.106
0.102
180.3
122.8
92.92
Min
-0.500
0.055
-0.500
-0.400
0.100
-0.400
-0.400
0.100
-0.400
-415.0
2.000
-415.0
Max
0.750
0.750
0.000
1.100
1.100
0.000
0.600
0.600
0.000
705.0
705.0
-1.000
N161
29132
189
170
19190
66124
190
150
40
μsur
-0.011
0.271∗∗∗
-0.256∗∗∗
-0.015
0.152∗∗∗
-0.125∗∗∗
-0.040∗∗
0.162∗∗∗
-0.166∗∗∗
-10.09
110.3∗∗∗
-112.2∗∗∗
Medsur
-0.060
0.180
-0.145
-0.100
0.100
-0.100
-0.100
0.100
-0.200
-31.00
90.00
-91.00
σsur
0.380
0.296
0.255
0.149
0.071
0.053
0.183
0.108
0.075
133.6
74.99
73.63
Nsur
7133
38128
5177
140
5486
146
6779
25
Table 3. GARCH model for excess returns: Surprise events
This table reports quasi-maximum likelihood (QML) estimates and robust t-statistics (Bollerslev andWooldridge,
1992), over a sample from 01/03/1986 to 02/14/2002 (4069 observations), for the GARCH(1,1) model of Eq. (1):
rit = μei + ρeirit−1 + γei (0) I
et (0)S
et (0) + εit
εit =psite
it eit|zt−1 ∼ N
¡0, hit
¢hit = ωei + αei
¡eit−1
¢2+ βeih
it−1,
where zt−1 denotes the information set, sit = 1 +P+1
k=−1 δei (k) I
et (k) |Set (k)|, rit is the daily continuously
compounded excess return on asset i (in percentage), and Set are actual standardized news. In Eq. (1), Iet (k) = 1
and Set (k) = Set if a surprise Federal Reserve rate change (e = fed), a surprise CPI (e = cpi), unemployment
(e = une), or payroll announcement (e = pay) was made on day t+ k and zero otherwise. A “ ∗ ”, “ ∗∗ ”, or
“ ∗∗∗ ” indicate significance at the 10%, 5%, or 1% level, respectively.
rnyxt rnaqt r5yt r10yt r30yt raaat rbaat
γfedi (0) -0.125 -0.135∗ -0.004 -0.015 0.018 0.002 -0.009-1.54 -1 .71 -0 .14 -0 .34 0.30 0.05 -0 .30
δfedi (−1) -0.134∗∗∗ -0.127∗∗∗ -0.049 -0.030 -0.026 -0.018 -0.027-3.91 -3 .81 -0 .79 -0 .47 -0.45 -0 .35 -0 .54
δfedi (0) 0.048 0.099 -0.067 0.021 0.070 0.007 0.0360.24 0.56 -1 .18 0.22 0.75 0.11 0.44
δfedi (1) 0.138 0.062 0.450∗∗∗ 0.129 0.021 0.075 0.1730.65 0.42 2.67 1.09 0.27 0.89 1.54
γcpii (0) 0.089 0.040 0.012 0.012 0.033 0.013 0.0221.28 0.63 0.67 0.40 0.71 0.59 0.89
δcpii (−1) -0.189∗∗∗ 0.002 0.157 0.119 0.039 -0.074 -0.121∗∗
-3 .74 0.03 1.51 1.18 0.44 -1 .04 -2 .10
δcpii (0) 0.988∗∗∗ 0.594∗∗∗ -0.178∗∗∗ -0.178∗∗∗ -0.154∗∗ -0.120 -0.1074.53 3.15 -2 .85 -2 .89 -2.31 -1 .62 -1 .55
δcpii (1) -0.104 -0.199∗∗∗ 0.177∗ 0.213∗∗ 0.224∗∗ 0.107 0.178∗
-1 .64 -3 .83 1.71 2.12 2.23 1.25 1.94
γunei (0) -0.092 -0.142∗∗ 0.012 0.011 -0.004 0.020 -0.013-1.57 -2 .44 0.52 0.29 -0.08 0.76 -0 .47
δunei (−1) -0.119∗∗∗ -0.098 -0.155∗∗∗ -0.166∗∗∗ -0.129∗∗∗ -0.137∗∗∗ -0.087-2.91 -1 .33 -4 .10 -5 .13 -3.38 -2 .96 -1 .51
δunei (0) 0.543∗∗∗ 0.352∗∗ -0.203∗∗∗ -0.183∗∗∗ -0.168∗∗∗ -0.160∗∗∗ -0.172∗∗∗
3.57 2.09 -14.86 -9 .25 -7.05 -4 .36 -6 .23
δunei (1) -0.091 -0.081 1.035∗∗∗ 0.912∗∗∗ 0.643∗∗∗ 0.624∗∗∗ 0.597∗∗∗
-1 .52 -0 .97 6.81 6.53 5.54 5.23 4.83
γpayi (0) -0.104 -0.066 0.016 0.036 0.038 0.013 0.018-1.51 -0 .91 0.73 1.00 0.71 0.49 0.67
δpayi (−1) -0.221∗∗∗ -0.182∗∗∗ -0.148∗∗∗ -0.123∗∗ -0.112∗∗ -0.104∗ -0.093-7.67 -4 .19 -3 .12 -2 .26 -2.12 -1 .69 -1 .62
δpayi (0) 0.853∗∗∗ 0.465∗∗ -0.249∗∗∗ -0.236∗∗∗ -0.210∗∗∗ -0.163∗∗∗ -0.198∗∗∗
4.38 2.46 -11.99 -7 .99 -6.40 -3 .13 -5 .35
δpayi (1) -0.170∗∗∗ -0.088 1.152∗∗∗ 0.943∗∗∗ 0.680∗∗∗ 0.621∗∗∗ 0.565∗∗∗
-3 .76 -1 .04 6.24 5.77 5.33 4.73 4.60
26
Table 4. DCC IMA model: Surprise events
This table reports intra and across-asset class averages of quasi-maximum likelihood (QML) estimates and of
their robust t-statistics (Bollerslev and Wooldridge, 1992) for the following DCC IMA model of Eq. (3):
ηitηjt = s
i∗t η
it−1η
jt−1 +
¡qijt − ηitη
jt
¢− λesi∗t
¡qijt−1 − ηit−1η
jt−1¢,
where si∗t = 1 +P+1
k=−1 dei (k) I
et (k) |Set (k)|, ηit =
εit√sith
it
is the daily standardized residual from the
GARCH(1,1) model of Eq. (1) for rit, the excess return on asset i, Set are actual standardized news, and
Iet (k) = 1 and Set (k) = S
et+k if a surprise Federal Reserve rate change (e = fed), a surprise CPI (e = cpi),
unemployment (e = une), or payroll announcement (e = pay) was made on day t+ k and zero otherwise. Eq.
(3) is estimated between 01/03/1986 and 02/14/2002, i.e., over 4069 observations. The Stock category is made
of i = nyx, naq; the Govt category is made of i = 5y, 10y, 30y; the Corp category is made of i = aaa, baa.
dfedi (−1) dfedi (0) dfedi (1)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.874 -0.881 -0.783 0.181 -0.320 -0.696 -0.274 -0.652 -1.111
-25.03 -16.35 -11.67 1.28 -8 .84 -12.42 -37.41 -15.65 -13.06
Govt -0.330 -0.912 -0.429 -0.685 -0.248 -0.012-7.31 -14.85 -7 .75 -28.32 -7 .73 -2 .51
Corp -0.151 -0.794 0.580-11.39 -38.15 4.05
dcpii (−1) dcpii (0) dcpii (1)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock 0.700 0.348 -1.555 0.564 -1.192 -1.072 -0.938 -0.763 -0.852
7.48 3.36 -25.14 6.57 -19.16 -18.29 -18.20 -12.55 -14.40
Govt -2.685 -1.152 -0.389 -0.466 -0.323 -0.657-22.64 -15.12 -11.96 -13.90 -7 .58 -17.92
Corp -0.106 -0.255 -0.808-2.46 -6 .92 -21.42
dunei (−1) dunei (0) dunei (1)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.206 -0.727 -0.844 -1.138 -1.033 -0.934 -1.267 -1.096 -0.992
-4.99 -14.50 -17.74 -37.20 -15.85 -17.29 -30.54 -20.94 -18.91
Govt -0.495 -1.077 -0.866 -1.504 0.091 -0.427-7.93 -22.84 -20.37 -30.78 1.41 -14.98
Corp -1.161 -1.700 -0.073-29.35 -40.82 -1 .39
dpayi (−1) dpayi (0) dpayi (1)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.948 -0.881 -0.746 -0.389 -0.935 -1.059 -0.314 -1.016 -1.055
-33.18 -11.35 -9.51 -24.25 -13.44 -13.89 -9 .16 -16.80 -15.02
Govt -0.545 -0.362 -1.191 -1.224 0.001 -0.032-13.29 -11.39 -27.96 -31.40 0.02 -0 .69
Corp -0.300 -1.294 -0.009-4.52 -34.37 -0 .17
27
Table 5. GARCH model for excess returns: Asymmetric impact
This table reports quasi-maximum likelihood (QML) estimates and robust t-statistics (Bollerslev andWooldridge,
1992), over a sample from 01/03/1986 to 02/14/2002 (4069 observations), for the GARCH(1,1) model of Eq. (4):
rit = μi + ρirit−1 + [γ
ei (0,+) I
et (0,+) + γei (0,−) Iet (0,−)]Set + εit
εit =
qsi±t e
it eit|zt−1 ∼ N
¡0, hit
¢hit = ωi + αi
¡eit−1
¢2+ βih
it−1,
where si±t = 1 +P+1
k=−1 δei (k,+) I
et (k,+) |Set (k,+)| +
P+1k=−1 δ
ei (k,−) Iet (k,−) |Set (k,−)|, zt−1
denotes the information set, rit is the daily continuously compounded excess return on asset i, Set are actual
standardized news, and Iet (k,±) and Set (k,±) are the surprisingly ”good” and ”bad” macroeconomic newsevent dummies defined in Section 5. A “ ∗ ”, “ ∗∗ ”, or “ ∗∗∗ ” indicate significance at the 10%, 5%, or 1% level,
respectively.
rnyxt rnaqt r5yt r10yt r30yt raaat rbaat
γfedi (0,+) -0.023 -0.026 0.004 -0.004 0.025 -0.004 -0.013-0.19 -0 .28 0.10 -0 .07 0.31 -0 .10 -0 .27
γfedi (0,−) -0.190∗ -0.246∗∗ -0.009 -0.017 0.005 0.013 -0.003-1.70 -2 .03 -0 .23 -0 .27 0.05 0.28 -0 .07
δfedi (−1,+) -0.084 0.007 -0.030 0.032 0.024 0.017 -0.011-0.34 0.04 -0 .28 0.27 0.29 0.24 -0 .16
δfedi (0,+) -0.046 0.037 0.390 0.244 0.069 0.101 0.205-0.10 0.15 1.27 1.07 0.65 0.85 1.18
δfedi (1,+) 0.326 0.018 -0.039 -0.036 0.011 -0.003 0.0830.76 0.10 -0 .34 -0 .35 0.13 -0 .04 0.62
δfedi (−1,−) -0.189∗∗∗ -0.193∗∗∗ -0.062 -0.076 -0.122∗∗ -0.073 -0.056-6.39 -8 .59 -0 .76 -0 .97 -1.91 -0 .92 -0 .68
δfedi (0,−) 0.304 0.166 -0.124∗∗ -0.057 0.129 -0.092 -0.0721.07 0.59 -2 .15 -0 .60 0.55 -1 .18 -0 .87
δfedi (1,−) -0.067 0.051 0.562∗∗∗ 0.219 0.058 0.274 0.280∗
-0 .46 0.24 2.72 1.36 0.31 1.63 1.65
γcpii (0,+) 0.132 0.094 -0.022 -0.032 -0.036 -0.022 -0.0201.57 1.19 -1 .01 -0 .87 -0.57 -0 .63 -0 .51
γcpii (0,−) 0.014 -0.039 0.044∗ 0.054 0.101 0.047 0.058∗
0.14 -0 .41 1.77 1.30 1.55 1.64 1.90
δcpii (−1,+) 0.076 0.188 0.510∗∗ 0.402∗∗ 0.183 -0.159∗ -0.225∗∗∗
0.61 1.27 2.53 2.19 1.19 -1 .85 -4 .12
δcpii (0,+) 1.039∗∗∗ 0.600∗∗ -0.293∗∗∗ -0.279∗∗∗ -0.251∗∗∗ -0.166∗ -0.0873.03 2.26 -6 .36 -5 .63 -4.45 -1 .67 -0 .81
δcpii (1,+) -0.277∗∗∗ -0.280∗∗∗ 0.307∗ 0.299∗ 0.382∗∗ 0.414∗∗ 0.372∗∗
-6 .87 -5 .57 1.75 1.93 2.34 2.48 2.28
δcpii (−1,−) -0.315∗∗∗ -0.143∗ -0.025 -0.028 -0.035 0.074 0.072-9.73 -1 .77 -0 .24 -0 .26 -0.33 0.62 0.64
δcpii (0,−) 0.809∗∗∗ 0.599∗∗∗ -0.166∗∗ -0.139 -0.106 -0.068 -0.142∗
3.39 2.65 -1 .99 -1 .47 -1.05 -0 .63 -1 .66
δcpii (1,−) 0.049 -0.143∗ 0.076 0.121 0.104 -0.148∗ -0.0070.43 -1.77 0.63 0.95 0.85 -1 .91 -0 .07
28
Table 5 (Continued).
rnyxt rnaqt r5yt r10yt r30yt raaat rbaat
γunei (0,+) -0.100 -0.111 -0.006 -0.021 -0.016 0.036 -0.016-0.96 -1 .06 -0 .13 -0 .31 -0 .17 0.77 -0.34
γunei (0,−) -0.092 -0.172∗∗ 0.007 0.009 -0.008 -0.006 -0.020-1.24 -2 .50 0.27 0.23 -0 .14 -0 .22 -0.68
δunei (−1,+) -0.111∗ -0.171∗∗∗ -0.146∗∗ -0.170∗∗∗ -0.119∗∗ -0.182∗∗∗ -0.082-1.74 -3 .86 -2 .17 -4 .32 -2 .48 -5 .55 -0.65
δunei (0,+) 0.603∗∗ 0.498∗ -0.194∗∗∗ -0.148∗∗∗ -0.117∗∗ -0.072 -0.144∗
2.34 1.87 -6 .93 -3 .28 -2 .35 -0 .62 -1.74
δunei (1,+) -0.012 0.032 1.224∗∗∗ 0.961∗∗∗ 0.578∗∗∗ 0.759∗∗∗ 0.766∗∗∗
-0 .10 0.19 5.28 4.78 3.92 3.20 3.71
δunei (−1,−) -0.221∗∗∗ 0.056 -0.063 0.000 0.017 0.170 0.061-3.22 0.40 -0 .70 0.00 0.16 1.42 0.56
δunei (0,−) 0.840∗∗∗ 0.134 -0.348∗∗∗ -0.341∗∗∗ -0.323∗∗∗ -0.345∗∗∗ -0.324∗∗∗
3.24 0.65 -15.05 -10.21 -12.98 -16.54 -11.34
δunei (1,−) -0.226∗∗∗ -0.133 1.095∗∗∗ 0.982∗∗∗ 0.808∗∗∗ 0.538∗∗∗ 0.614∗∗∗
-2 .77 -1 .23 4.89 3.44 4.21 3.44 3.80
γpayi (0,+) -0.194∗ -0.112 -0.042 -0.032 -0.040 -0.008 -0.025-1.85 -1 .17 -1 .24 -0 .58 -0 .50 -0 .22 -0.67
γpayi (0,−) -0.038 -0.029 0.066∗∗ 0.091∗ 0.096 0.025 0.057-0.41 -0 .25 2.37 1.88 1.33 0.62 1.57
δpayi (−1,+) -0.212∗∗∗ -0.167∗∗∗ -0.175∗∗∗ -0.151∗∗ -0.103 -0.028 -0.130∗∗
-5 .74 -2 .99 -3 .43 -2 .22 -1 .43 -0 .28 -2.12
δpayi (0,+) 0.643∗∗∗ 0.206 -0.238∗∗∗ -0.235∗∗∗ -0.205∗∗∗ -0.217∗∗∗ -0.196∗∗∗
2.91 1.00 -9 .05 -7 .93 -5 .37 -6 .35 -4.81
δpayi (1,+) -0.154∗∗ -0.040 1.101∗∗∗ 0.970∗∗∗ 0.589∗∗∗ 0.430∗∗∗ 0.448∗∗∗
-2 .36 -0 .30 4.70 4.32 3.75 2.78 3.18
δpayi (−1,−) -0.225∗∗∗ -0.225∗∗∗ -0.040 -0.112 -0.121 -0.158∗∗ -0.024-3.72 -2 .95 -0 .38 -1 .38 -1 .49 -2 .36 -0.22
δpayi (0,−) 1.143∗∗∗ 0.832∗∗ -0.305∗∗∗ -0.210∗∗∗ -0.215∗∗∗ -0.063 -0.215∗∗∗
3.50 2.46 -7 .46 -3 .35 -3 .59 -0 .62 -3.27
δpayi (1,−) -0.218∗∗∗ -0.128 1.175∗∗∗ 0.857∗∗∗ 0.766∗∗∗ 0.661∗∗∗ 0.644∗∗∗
-3 .23 -1 .19 4.68 4.18 4.08 3.45 3.51
29
Table 6. DCC IMA model: Asymmetric covariance shifts
This table reports intra and across-asset class averages of quasi-maximum likelihood (QML) estimates and of
their robust t-statistics (Bollerslev and Wooldridge, 1992) for the DCC IMA model of Eq. (5):
ηitηjt = s
i±∗t ηit−1η
jt−1 +
¡qijt − ηitη
jt
¢− λesi±∗t
¡qijt−1 − ηit−1η
jt−1¢
where si±∗t = 1 +P+1
k=−1 dei (k,+) I
et (k,+) |Set (k,+)| +
P+1k=−1 d
ei (k,−) Iet (k,−) |Set (k,−)|, ηit =
εit√sith
it
is the daily standardized residual from the GARCH(1,1) model of Eq. (1) for rit, the excess return
on asset i, Set are actual standardized news, and Iet (k,±) and Set (k,±) are the surprisingly ”good” and
”bad” macroeconomic news event dummies defined in Section 5. Eq. (5) is estimated between 01/03/1986 and
02/14/2002, i.e., over 4069 observations. The Stock category is made of i = nyx, naq; the Govt category is
made of i = 5y, 10y, 30y; the Corp category is made of i = aaa, baa.
dfedi (−1,+) dfedi (0,+) dfedi (1,+)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.999 -0.772 -0.718 -0.194 -0.863 -0.881 -1.270 -1.251 -0.476
-17.75 -9 .91 -10.22 -6 .24 -12.31 -11.33 -50.09 -18.71 -9 .71
Govt -0.150 -0.960 -0.750 -0.601 -0.303 -0.314-8.82 -10.43 -8 .67 -5 .92 -9 .68 -10.13
Corp -0.154 0.072 -0.251-13.73 1.14 -13.86
dfedi (−1,−) dfedi (0,−) dfedi (1,−)Stock Govt Corp Stock Govt Corp Stock Govt Corp
Stock -0.902 -0.989 -0.936 0.170 -0.258 -0.652 -0.265 -1.105 -0.938-45.29 -23.54 -19.84 1.19 -6 .58 -11.36 -54.25 -23.39 -21.29
Govt -0.553 -0.733 -0.261 -0.743 -0.202 0.132-7.89 -11.78 -6 .60 -45.61 -8 .08 -0 .85
Corp -0.703 -0.803 0.712-11.08 -47.49 3.47
dcpii (−1,+) dcpii (0,+) dcpii (1,+)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock 0.606 0.441 -1.035 -0.105 -0.629 -0.751 -0.277 -0.649 -0.758
5.28 3.07 -11.40 -1 .38 -7 .49 -9 .35 -5 .07 -9 .23 -11.32
Govt -3.375 -1.587 -0.392 -0.445 -1.028 -0.595-12.73 -12.23 -8 .42 -8 .71 -11.03 -14.43
Corp -2.751 -0.263 -0.081-14.01 -5 .35 -1 .30
dcpii (−1,−) dcpii (0,−) dcpii (1,−)Stock Govt Corp Stock Govt Corp Stock Govt Corp
Stock -0.591 -0.665 -1.058 0.805 -1.077 -0.920 -1.088 -0.918 -0.984-37.93 -11.73 -8.62 6.21 -16.75 -9 .50 -16.20 -10.27 -11.82
Govt -0.656 -0.928 -0.327 -0.363 -0.209 -0.815-9.89 -11.48 -6 .22 -5 .23 -3 .39 -12.62
Corp -1.156 -0.257 -1.017-19.08 -4 .22 -16.72
30
Table 6 (Continued).
dunei (−1,+) dunei (0,+) dunei (1,+)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.172 -0.639 -0.828 -0.118 -1.112 -1.257 -1.077 -1.124 -0.957
-3.21 -12.52 -14.46 -1 .89 -14.22 -19.23 -21.75 -22.06 -14.75
Govt -0.850 -1.298 -1.672 -1.394 0.020 -0.438-18.64 -21.10 -21.29 -19.66 0.23 -10.04
Corp -1.165 -1.757 -0.206-33.12 -32.51 -4 .44
dunei (−1,−) dunei (0,−) dunei (1,−)Stock Govt Corp Stock Govt Corp Stock Govt Corp
Stock -0.252 -0.737 -0.946 -0.089 -0.984 -1.069 -1.356 -1.097 -1.107-4.32 -12.15 -15.57 -1 .39 -18.71 -22.51 -30.23 -18.82 -19.33
Govt -0.680 -1.106 -0.752 -0.728 0.071 0.043-13.47 -18.27 -28.11 -28.22 0.88 0.57
Corp -0.654 -0.641 -2.703-16.45 -33.93 -23.58
dpayi (−1,+) dpayi (0,+) dpayi (1,+)
Stock Govt Corp Stock Govt Corp Stock Govt CorpStock -0.178 -0.888 -0.857 -1.107 -0.857 -1.048 -0.447 -0.821 -0.664
-7.02 -13.67 -11.88 -30.25 -8 .97 -12.03 -7 .98 -14.67 -13.07
Govt -0.922 -0.708 -1.161 -1.222 0.233 0.279-23.39 -15.87 -19.37 -24.75 1.94 2.19
Corp -0.175 -1.271 0.356-3.26 -28.64 2.69
dpayi (−1,−) dpayi (0,−) dpayi (1,−)Stock Govt Corp Stock Govt Corp Stock Govt Corp
Stock -1.013 -0.736 -0.583 -0.291 -0.924 -0.986 -0.167 -1.137 -0.986-29.96 -12.37 -13.26 -5 .63 -11.46 -12.32 -2 .63 -13.88 -12.45
Govt -0.476 -0.202 -1.267 -1.230 -0.156 -0.238-17.01 -13.82 -22.70 -23.42 -3 .06 -5 .35
Corp 0.330 -1.246 -0.2051.87 -23.26 -3 .92
31